Modeling of content is currently performed at a very coarse level. For example, content modeling currently is performed based on the web page that a user loaded, for example. Some video devices track fine grained navigation, but it is only coarsely linked to the specific content that is being accessed. For example, Tivo tracks fast forward and reverse choices and ties that to the portions of ads or shows that are watched or skipped, but there are no deeper semantics. Similarly, some email programs track the emails that were accessed in order to support prioritization, but do not track which portions of those emails being accessed.